International Capital Mobility in African Countries:

A Panel Cointegration Analysis

Chrysost BANGAKE
Laboratoire d’Economie d’Orléans (LEO), Université d’Orléans
Faculté de Droit, d’Economie et de Gestion. Rue de Blois BP : 6739. 45067 Orléans Cedex 2
E.mail : chrysost.bangake@univ-orleans.fr

Jude C. EGGOH1
Laboratoire d’Economie d’Orléans (LEO), Université d’Orléans
Faculté de Droit, d’Economie et de Gestion. Rue de Blois BP : 6739. 45067 Orléans Cedex 2
E.mail : comlanvi-jude.eggoh@univ-orleans.fr

Abstract

This paper investigates the Feldstein-Horioka coefficients for 37 African countries using the
recently developed Pool Mean Group (PMG), Fully Modified OLS (FMOLS) and Dynamic
OLS (DOLS) panel cointegration techniques. The empirical findings reported in the paper
reveal that savings and investment are nonstationary and cointegrated series. The estimated
coefficients using FMOLS, DOLS and PMG are 0.38, 0.58 and 0.36, respectively for the
sample as a whole for the period 1970-2006. These results confirm the previous studies that
capital was relatively mobile in African countries. In addition, our study shows that there are
marked differences in saving-retention coefficients for different country groups in Africa
(Franc CFA zone countries and non Franc CFA zone countries; oil producing and non-oil
producing countries; civil law and common law countries).

Key Words: Feldstein-Horioka coefficient; panel unit root test; PMG panel cointegration; FMOLS panel
cointegration; DOLS panel integration; Africa.

1
Corresponding author : Jude Eggoh, Laboratoire d’Economie d’Orléans, Faculté de Droit, d’Economie et de
Gestion. Rue de Blois, B.P 6739-45067 Orléans Cedex 2 (France), e-mail : comlanvi-jude.eggoh@univ-
orleans.fr

1
1. Introduction

Increasing capital mobility across countries is an important phenomenon for economic policy
makers and firms. It is generally admitted that most African countries keep significant legal
restrictions over capital movements and have limited financial market linkages with world
economy. Such a situation has contributed to weak economic growth, a relatively low saving
rate and capital flight from the region (Collier and Gunning, 1999). In the economics
literature, there are several ways to investigate the degree of international capital mobility2.
One test proposed by Feldstein and Horioka (1980, hereafter FH) for capital mobility is to
examine the relationship between saving and investment. According to these authors, in the
absence of capital mobility, domestic saving and investment are highly correlated since
investment is financed by domestic saving. Since the work of FH, many economists have
studied the relationship between savings and investment3.

However many of these studies focused on cross section data and times series aspects of data.
Savings and investment usually turn out to be nonstationary. In the unit-root literature, it is
argued that the widespread failure of hypothesis testing in relatively short series may be
accounted for by the low power of conventional univariate unit root tests against persistent
alternatives, typically for sample sizes that occur in practice. Further, the traditional
cointegration technique has also the problem of low power.

In this paper, we make a contribution to the debate on the Feldstein-Horioka puzzle in African
countries in three ways. First, we attempt to employ a battery of new heterogeneous panel unit
root and cointegration tests. This method avoids problems of low power associated with the
traditional unit root and cointegration tests. Another advantage of using panel is that it reduces
collinearity between regressors.

To the best of our knowledge, only Adedeji and Thornton (2006) have examined the
Feldstein-Horioka puzzle using panel cointegration on African countries but our study differs
from theirs in more ways than one. Their study is limited to six African countries and does not
consider different groups of countries in Africa. Furthermore, the long run relationship is
analysed using only FMOLS and DOLS estimators.

The second contribution of this study is the use of Pooled Mean Group (PMG) estimator
proposed by Pesaran, Shin and Smith (1999). The Pool Mean group is an intermediate
estimator that allows the short-term parameters to differ between groups while imposing
equality of the long-term coefficients between groups4.

The third contribution lies in the sample that considers groups of counties in Africa (Franc
CFA zone countries and non Franc CFA zone countries; oil producing and non-oil producing
countries; civil law and common law countries).

2
The presence of capital mobility is tested alternatively by using the saving correlation, uncovered interest parity
condition, and finally the consumption smoothing approach to the current account. See Obstfeld (1993) for more
detail.
3
See Coakley and al. (1998) for a survey.
4
Pesaran et al. (1999) argue that there are often good reasons to expect the same long-run equilibrium
relationships across countries, due to budget or solvency constraints, arbitrage conditions, or common
technologies influencing all groups in a similar way. The reasons for assuming that short-run dynamics and error
variance should be the same tend to be less compelling.

2
The paper is organized as follows. The next section briefly reviews the literature on the
savings and investment relationship. Section 3 introduces the empirical methodology. The
data and empirical results are presented and interpreted in section 4. Section 5 concludes and
presents some policy implications.

2. Brief literature review

Feldstein and Horioka (1980) propose assessing the degree of capital mobility by measuring
the correlation between saving and development. They estimate the following cross-section
regression:
I S
        i , (1)
 Y i  Y i
I S
where   and   are respectively the saving and investment rates of country i ,  is the
 Y i  Y i
savings-retention coefficient and  is the error term. For a small, open economy where capital
is perfectly mobile internationally,  should be close to zero. If  equals zero, then there is
no relationship between saving and investment. Feldstein and Horioka (1980) suppose that if
 is large, however, capital is considered immobile internationally. For example, if  equals
one, then all additional saving goes to finance domestic investment.

Using the averaged cross-section data across 16 OECD countries for the period 1960-1974,
Feldstein and Horioka (1980), find the estimated coefficient of  ranging from 0.87 to 0.91.
They conclude that almost 90 percent of domestic saving remains within a country to finance
domestic investment. Therefore, capital is not internationally mobile, in contradiction with the
belief that the industrial countries had few barriers to capital movements. The high correlation
between savings and investment has been known as the Feldstein-Horioka puzzle.

The finding by Feldstein and Horioka (1980) that national saving and domestic investment are
highly positively correlated has generated a lot of studies in this area. We will do selective
surveys on the savings and investment relation using panel data.

Krol (1996) argues that the Feldstein-Horioka puzzle is related to estimated technique and
reported estimates of the saving-investment correlation based on panel regressions. He finds
that the saving-retention coefficients are much lower than those obtained by Feldstein and
Horioka (1980) and many others. Krol postulates that the large estimates reported in earlier
works are due to intertemporal budget constraints, which require that current account deficits
should not be allowed to persist infinitely. By definition, the current account balance of a
country in any one period equals the difference between investment and saving. Because the
current account must average zero over time, savings and investments move hand in hand in
the long term. Cross- sectional investment regressions using time-averaged data will tend to
reject international capital mobility. Using panel data for 21 OECD countries over the 1962-
1990 period and controlling for business cycle effects, Krol’s point estimate for saving-
investment correlation  is only 0.20.

Coiteux and Olivier (2000), Jansen (2000) criticize Krol’s low estimates as resulting from the
inclusion of Luxembourg in his fixed-effect model, and show that the exclusion of
Luxembourg would result in larger estimates. Panel estimation reduces the correlation’s only
by about 0.12. Ho (2002) reexamines the Luxembourg problem by applying Kao and

3
Chiang’s (2000) DOLS and FMOLS estimators to non stationary panel data of 20 OECD
countries. He shows that the inclusion or exclusion of Luxembourg does not affect the
estimation results. His saving-retention coefficients tend to be low.

Coakley and al. (2003) reassesses the Feldstein-Horioka puzzle in a nonstationary framework
for a sample of 12 OECD economies over the period 1980-2000. The group-mean OLS panel
estimator gives a slope coefficient estimate which is insignificantly different from zero. This
supports the hypothesis of long run capital mobility.

Recently, Kim and al. (2005) investigates the saving-investment relationship by applying
FMOLS and DOLS panel cointegration techniques for 11 Asian countries. The FH
coefficients using FMOLS and DOLS are 0.39 and 0.42, respectively, for the period 1980-
1998, whereas they are 0.58 and 0.79 for 1960-1979. The small coefficients suggest that
capital mobility increased in Asian countries in the 1980s and 1990s.

While there have been increasing empirical studies on OECD countries, there is a limited
number of empirical attempts to verify the presence of capital mobility using the FH approach
for African countries. To the best of our knowledge, there are only three papers that deal with
capital mobility using panel data in Africa5. Works by Payne and Kamazawa (2005), and De
Wet and Van Eyden (2005) use fixed and random effects to examine the relation saving-
investment relationships in sub-Saharan African countries over the period 1980-2000. The
empirical evidence suggests the presence of capital mobility in Africa. Adedeji and Thornton
(2006) is an interesting paper using panel cointegration. They applied to data for six African
countries panel cointegration techniques to test the FH approach. Their main result is that
capital was relatively mobile in the African countries during 1970-2000, with estimated
savings-retention ratios of 0.73 (FMOLS), 0.45 (DOLS), 0.51 (DOLS with heterogeneity) and
0.39 (DOLS with cross-sectional dependence effects). Unfortunately, as mentioned above, the
study by Adedeji and Thornton does not undertake various country groups in Africa and panel
cointegration tests using PMG estimator. This paper tries to fill this gap.

3. Empirical methodology

3.1 Panel unit root tests

Before proceeding to cointegration techniques, we need to verify that all variables are
integrated to the same order. In doing so, we have used first generation tests of panel unit root
due to Im, Pesaran and Shin (2003) and Maddala and Wu (1999)6 and second generation test
of panel unit root of Pesaran (2005). These tests are less restrictive and more powerful
compared to the tests developed by Levin and Lin (1993, 2002) 7 , which don’t allow for
heterogeneity in the autoregressive coefficient. The tests proposed by IPS permit to solve
Levin and Lin’s serial correlation problem by assuming heterogeneity between units in a

5
For studies using cross-section, times series techniques and other methodologies on developing countries
including African countries, see Isakson (2001), Rocha (2000), Hussein and Mello (1999), Mamingi (1997),
Montiel (1994), Haque and Montiel (1990), and Dooley, Frankel and Mathieson (1987).
6
Henceforth, IPS for Im, Pesaran and Shin, and MW for Maddala and Wu.
7
For a useful survey on panel unit root tests, see Hurlin and Mignon (2005) and Banerjee (1999).

4
dynamic panel framework. The basic equation for the panel unit root tests for IPS is as
follows:
p
yi , t   i   i yi , t 1  ij yi ,t  j   i , t ; i  1, 2,...,N ; t  1, 2,...,T , (2)
j 1

where yi , t stands for each variable under consideration in our model,  i is the individual
fixed effect and p is selected to make the residuals uncorrelated over time. The null
hypothesis is that  i  0 for all i versus the alternative hypothesis is that  i  0 for some
i  1,..., N1 and  i  0 for i  N1  1,...,N .
The IPS statistic is based on averaging individual ADF statistics and can be written as follows:
1 N
t   (t iT ), (3)
N i 1
where tiT is the ADF t-statistic for country i based on the country-specific ADF regression, as
in Eq (1). IPS show that under the null hypothesis of non stationary, the t statistic follows the
standard normal distribution asymptotically. The standardized statistic t IPS is expressed as:

 1 N 
n  t   E tiT | i  0 
t IPS   N i 1 . (4)
N
1
Var tiT | i  0
N i 1

Maddala and Wu (1999) argue that while Im et al.’s tests relax the assumption of
homogeneity of the root across the units, several difficulties still remain. They suggest the use
of a Fisher type test which is based on combining the p-values,  i of the test-statistic for a
unit root in each cross-sectional unit. The MW test statistic  is given by:

N
  2 ln  i . (5)
i 1

The MW test statistic is distributed as Chi square with 2N degrees of freedom under the
hypothesis of cross-sectional independence.

According to Breitung (1999), IPS’s test is not powerful when individual trends are included.
This test is sensitive to the specification of deterministic trends compared to IPS’s test. The
MW test has the advantage that its value does not depend on different lag lengths in the
individual ADF regressions. Furthermore, Maddala and Wu (1999) found that MW’s test is
superior compared to IPS’s test.

Both the tests (IPS and MW) have the drawback to suppose that the cross-sections are
independent; the same assumption is made in all first generation of panel unit root. However,
it has been pointed out in the literature that cross section dependence arises due to unobserved
common factors, externalities, regional and macroeconomic linkages, and unaccounted
residual interdependence. Recently, some new panel unit root tests have emerged and address
the question of the dependence and correlation given the prevalence of macroeconomic
dynamics and linkages. These tests are called the second generation panel unit root tests. The
well-known second generation test that is considered in this paper is the Pesaran’s CIPS test
(2005). In order to formulate a panel unit root test with cross-sectional dependence, Pesaran

5
(2005) considers the following Cross-Sectional Augmented Dickey-Fuller (CADF) regression,
estimated the OLS method for the ith cross-section in the panel:

k k
yit  i  i yi ,t 1   i yt 1   ij yi ,t  j   yi ,t  j   it (6)
j 0 j 0

1 N 1 N
where, yt 1     yi ,t 1 , yt     yit , and ti  N , T  is the t-statistic of the estimate of
 N  i 1  N  i 1
i in the above equation used for computing the individual ADF statistics. More preciously,
Pesaran proposed the following test CIPS statistic that is based on the average of individual
CADF statistics as follows:

1 N
CIPS     ti  N , T  . (7)
 N  i 1

The critical values for CIPS for various deterministic terms are tabulated by Pesaran (2005).

3.2 Panel cointegration tests

Once the order of stationary has been defined, we would apply Predroni’s cointegration test
methodology. Indeed, like the IPS and MW panel unit root, the panel cointegration tests
proposed by Pedroni (1999) also take in account heterogeneity by using specific parameters
which are allowed to vary across individual members of the sample. Taking into account such
heterogeneity constitutes an advantage because it is unrealistic to assume that the vectors of
cointegration are identical from an individual to another for the panel.

The implementation of Pedroni’s cointegration test requires estimating first the following long
run relationship:

yit   i   i t  1i x1, it   2i x2, it  ...   Mi xM , it   it (8)

for i  1,...,N ; t  1,...,T ; m  1,...,M

where N refers to the numbers of individual members in the panel; T refers to the number of
observation over time; M refers to the number of exogenous variables. The structure of
estimated residuals is follows:
ˆit  ˆ iˆit1  uˆit . (9)

Pedroni has proposed seven different statistics to test panel data cointegration. Out of these
seven statistics, four are based on pooling, what is referred to as the “Within” dimension and
the last three are based on the “Between” dimension. Both kinds of tests focus on the null
hypothesis of no cointegration. However the distinction comes from the specification of the
alternative hypothesis. For the tests based on “Within”, the alternative hypothesis is
i    1 for all i, while concerning the last three test statistics which are based on the
“Between” dimension, the alternative hypothesis is i  1 , for all i.

6
The finite sample distribution for the seven statistics has been tabulated by Pedroni via Monte
Carlo simulations. The calculated statistic tests must be smaller than the tabulated critical
value to reject the null hypothesis of absence of cointegration.

3.3 Panel cointegration estimation

Although Pedroni’s methodology allows us to test the presence of cointegration, it could not
provide estimation of long-run relationship. For panel framework, in presence of cointegration,
several estimators are proposed: OLS, Fully Modified OLS (FMOLS) 8 , dynamic OLS
(DOLS), and Pooled Mean Group (PMG). Chen, McCoskey and Kao (1999) analysed the
proprieties of the OLS estimator9 and found that the bias-corrected OLS estimator does not
improve over the OLS estimator in general. These results suggest that alternatives, such as the
FMOLS estimator or the DOLS estimator may be more promising in cointegrated panel
regressions. However, Kao and Chiang (2000) showed that both the OLS and Fully Modified
OLS (FMOLS) exhibit small sample bias and that the DOLS estimator appears to outperform
both estimators10.

In this paper, we consider three estimators with error correction: Fully Modified OLS
(FMOLS), dynamic OLS (DOLS) and Pool Mean Group (PMG) to empirically examine the
validity of the Feldstein-Harioka puzzle in African countries.

3.3.1. The Fully Modified OLS (FMOLS) and Dynamic OLS (DOLS) estimators

The Fully Modified OLS (FMOLS) and Dynamic OLS (DOLS) methodologies are proposed
by Kao and Chiang (2000) to estimate the long-run cointegration vector, for non-stationary
panels. These estimators correct the standard pooled OLS for serial correlation and
endogeneity of regressors that are normally present in long-run relationship.

Let us consider the following fixed effect panel regression:

yit  i  xit   uit , i  1,...,N , t  1,...,T , (10)
where yit is a matrix 1,1 ,  is a vector of slopes k ,1 dimension,  i is individual fixed
effect, uit are the stationary disturbance terms. It is assumed that xit k ,1 vector are integrated
processes of order one for all i , where:
xit  xit1   it . (11)
Under these specifications, (Eq. 9) describes a system of cointegrated regressions, i.e. yit is
cointegrated with xit . By examining the limiting distribution of the FMOLS and DOLS
estimators in co-integrated regressions, Kao and Chiang (2000) show that they are
asymptotically normal. The FMOLS estimator is constructed by making corrections for
endogeneity and serial correlation to the OLS estimator and is defined as:

8
The FMOLS is popular in conventional time series econometrics, for it is believed to eliminate endogeneity in
the regressors and serial correlation in the errors.
9
Following proprieties are examining by Chen et al. (1999): the finite sample proprieties of the OLS estimator,
the t-statistic, the bias-corrected OLS estimator, and the bias-corrected t-statistic.
10
See Kao and Chiang (2000) for more discussions on the advantages of these estimators.

7
 1
N T  N  T  
ˆ FM     xit  xi       xit  xi  yˆit  T ˆ   , (12)
 i 1 t 1   i 1  t 1 

where ˆ 

is the serial correlation correction term and yˆit is the transformed variable of yit to
achieve the endogeneity correction. The serial correlation and the endogeneity can also be
corrected by using DOLS estimator. The DOLS is an extension of Stock and Watson’s (1993)
estimator. In order to obtain an unbiased estimator of the long-run parameters, DOLS
estimator uses parametric adjustment to the errors by including the past and the future values
of the differenced I(1) regressors. The dynamic OLS estimator is obtained from the following
equation:

j  q2
yit  i  xit    c x
j  q1
ij i ,t  j  vit . (13)

where cij is the coefficient of a lead or lag of first differenced explanatory variables. The
estimated coefficient of DOLS is given by:

1
 T
N
  T 
ˆ DOLS     zit zit    zit yˆit  (14)
i 1  t 1   t 1 

where zit   xit  xi , xi ,t  q ,..., xi ,t  q  is 2  q  1 1 vector of regressors.

3.3.2 The Pooled Mean Group (PMG) by Pesaran et al. (1999)

Our final step consists in using alternative methodology, the Pooled Mean Group (PMG) to
estimate the cointegration relationship in Feldstein and Horioka puzzle in order to test the
robustness of the previous results. There are two estimation methods commonly used with
dynamic panel data models. The firs consist of averaging separate estimates for each group in
the panel. According to Pesaran and Smith (1995), the mean group estimator provides
consistent estimates of the parameter averages. It allows the parameters to be freely
independent across groups and does not consider potential homogeneity between groups. The
second method is the usual pooled method; examples are the random effects, fixed effects,
and GMM methods. These models force the parameters to be identical across groups, but the
intercept can differ between groups. GMM estimations of dynamic panel could lead to
inconsistent and misleading long-term coefficients, a possible problem that is exacerbated
when the period is broad (Pesaran, Shin, and Smith, 1999).

The PMG is an intermediate estimator because it involves both pooling and averaging. One
advantage of the PMG over the FMOLS and DOLS models is that it can allow the short-run
dynamic specification to differ from country to country while the long run coefficients are
constrained to be the same.
The Autoregressive Distributed Lag ARDL ( p, q, q,....,q) model proposed by Pesaran et al.
(1999) is:
p q
yit   ij yi , t  j   ij xi , t  j   i   it . (15)
j 1 j 0

8
The cross-section units (countries) are denoted by i  1, 2,..., N , t  1, 2,...,T represent time
periods, xit k ,1 is a vector of explanatory variables (regressors) for country i,  i represent
the fixed effect, ij the coefficient of the lagged dependent variables, and  ij are k  1
coefficient vectors.

It is convenient to work with the following re-parameterization of (15).

p 1 q 1
yit  i yi , t 1   i xit   *ij yi , t  j   ij* xi , t  j  i   it , (16)
j 1 j 0

where i  1  pj1 ij  ,  i  qj0  ij , *ij   mp  j 1 im , j  1, 2,..., p  1 and
 ij*   qm j 1 im , j  1, 2,...,q  1 .

Pesaran et al. (1999) assume that the ARDL ( p, q, q,....,q) model is stable if the roots of the
following equation 1   pj1 ij z j  0 lie outside the unit circle. This assumption ensures
that i  0 , and hence there exists a long-run relationship between y it and xit defined by
yit    i
i 
xit   it where it is a stationary process and the long-run coefficient  i   ii  
are the same across the group.

4. Data and empirical results

4.1. Data

The data are taken from the World Development Indicators (WDI, 2008) CD-ROM for 37
African countries for the period 1970-2006. Following the original study of Feldstein and
Horioka, savings is defined as gross domestic savings as a percentage of GDP while
investment is measured by gross fixed capital formation divided by GDP 11 . The data are
summarised in table 1 (in appendix) which shows marked differences between savings and
investment ratios within and across countries.

4.2. The unit root tests

Table 2 reports the outcome for the global sample of three panel unit root tests: IPS (2003),
MW (1999) and Pesaran (2003). It shows that the null hypothesis of the unit roots for the
panel data for the investment and savings series cannot be rejected in level. However, this
hypothesis is rejected when series are in first differences. These results strongly indicate that
the variables in level are non-stationary and stationary in first-differences. The same issues
are obtained for the panels of country groups of our sample: Franc CFA zone countries and
non Franc CFA zone countries (table 3); oil producing and non-oil producing countries (table
4); civil law and common law countries (table 5). Therefore, we can implement a test for
panel cointegration between savings and investment.

11
Bayoumi (1990) and Sinha and Sinha (2004) suggest the use of gross fixed capital formation, since it excludes
the procyclical inventories component that may lead to spurious correlations with savings.

9
Table 2: Panel unit root for saving and investment ratios, 1970-2006 (global sample)

Im et al. (2003) Maddala and Wu (1999) Pesaran (2005)

Statistic P-values Statistic P-values Statistic P-values
All countries (37)
 I / Y it -1.547 0.148 59.477 0.890 -1.421 0.988
 I / Y  it
-4.026 0.000 488.481 0.000 -3.913 0.000
 S / Y it -1.384 0.489 63.166 0.811 -1.599 0.866
  S / Y it -4.149 0.000 544.287 0.000 -3.766 0.000

Table 3: Panel unit root for saving and investment ratios, 1970-2006 (FCFA zone vs non
FCFA zone countries)
Im et al. (2003) Maddala and Wu (1999) Pesaran (2005)
Statistic P-values Statistic P-values Statistic P-values
FCFA zone countries (13)
I /Y  it
-1.595 0.212 24.948 0.521 -1.366 0.939
  I / Y it -4.077 0.000 184.512 0.000 -4.119 0.000
 S / Y it -1.674 0.137 20.578 0.763 -1.652 0.674
  S / Y it -4.326 0.000 216.767 0.000 -4.036 0.000
Non FCFA zone countries (24)
 I / Y it -1.581 0.155 34.528 0.927 -1.429 0.962
 I / Y  it
-4.015 0.000 303.969 0.000 -3.717 0.000
 S / Y it -1.358 0.544 35.047 0.918 -1.238 0.997
  S / Y it -4.027 0.000 327.520 0.000 -3.874 0.000

Table 4: Panel unit root for saving and investment ratios, 1970-2006 (Oil-producing vs.
non oil-producing countries)
Im et al. (2003) Maddala and Wu (1999) Pesaran (2005)
Statistic P-values Statistic P-values Statistic P-values
Oil-producing countries (13)
I /Y  it
-1.564 0.247 33.193 0.156 -1.731 0.560
  I / Y it -3.983 0.000 156.878 0.000 -3.681 0.000
 S / Y it -1.087 0.862 13.215 0.982 -1.416 0.913
  S / Y it -4.114 0.000 184.274 0.000 -3.709 0.000
Non Oil-producing countries (24)
 I / Y it -1.487 0.294 26.283 0.995 -1.510 0.912
 I / Y  it
-4.012 0.000 331.602 0.000 -3.810 0.000
 S / Y it -1.483 0.302 42.410 0.700 -1.151 0.999
  S / Y it -4.158 0.000 360.013 0.000 -3.799 0.000

10
Table 5: Panel unit root for saving and investment ratios, 1970-2006 (civil law vs
common law countries)
Im et al. (2003) Maddala and Wu (1999) Pesaran (2005)
Statistic P-values Statistic P-values Statistic P-values
Civil law countries (20)
I /Y  it
-1.480 0.322 35.583 0.669 -1.610 0.776
  I / Y it -4.052 0.000 273.646 0.000 -3.948 0.000
 S / Y it -1.244 0.735 24.456 0.974 -1.434 0.945
  S / Y it -4.300 0.000 320.241 0.000 -3.904 0.000
Common law countries (16)
 I / Y it -1.671 0.115 19.893 0.953 -1.279 0.982
 I / Y  it
-3.876 0.000 191.575 0.000 -3.871 0.000
 S / Y it -1.509 0.297 29.348 0.601 -1.081 0.998
  S / Y it -3.986 0.000 211.272 0.000 -3.792 0.000

4.3. Panel cointegration tests

Table 6 shows the outcomes of Pedroni’s (1999) cointegration tests between the investment
and savings rates. We use four within-group tests and three between-group tests to check
whether the panel data are cointegrated. The columns labeled within-dimension contain the
computed value of the statistics based on estimators that pool the autoregressive coefficient
across different countries for the unit root tests on the estimated residuals. The columns
labeled between-dimension report the computed value of the statistics based on estimators
that average individually estimated coefficients for each country. Except the v-statistic test,
the results of the within-group test and the between-group tests show that the null hypothesis
of no cointegration can not be rejected at the 1% significant level. Therefore, the ratios of
saving and investment are cointegrated for the panel of all countries and for the panels of
country groups.

The presence of a long-run relationship between investment and saving rate in the panel of
African countries are economically meaningful in that it suggests that these countries meet the
long-run solvency condition. Having found that there exists a cointegrating link between the
two variables (ratios of saving and investment), it is convenient that the savings-retention
coefficient be estimated using a panel cointegrating estimator. In this paper, we choose to
employ a several panel cointegrating estimators: the Fully Modified OLS, the Dynamic OLS
and the Pooled Mean Group.

11
Table 6: Pedroni Panel cointegration test results, 1970-2006
Within-dimension (panel) Between-dimension (group)
v -Stat  -Stat PP-Stat ADF-Stat  -Stat PP-Stat ADF-Stat
*** *** *** *** ***
All 0.538 -4.836 -6.282 -7.282 -2.659 -5.438 -6.032***
*** *** *** ** ***
FCFA 0.646 -3.601 -4.650 -6.151 -2.289 -4.260 -5.001***
*** *** *** ** ***
NonCFA 0.367 -3.766 -4.803 -5.178 -1.868 -4.013 -4.771***
Oil 1.076 -2.782*** -3.738*** -5.613*** -1.722** -3.543*** -5.285***
*** *** *** **
Non Oil -0.063 -3.858 -4.804 -4.847 -1.750 -3.858*** -3.730***
French 1.564* -5.093*** -6.274*** -7.623*** -2.389*** -4.686*** -5.835***
* ***
English -0.533 -0.875 -1.375 -1.731 -0.415 -1.995** -2.729***
Notes: Results with a trend and time-dummies. The test statistics are normalized so that the asymptotic
distribution is standard normal. *, **, *** indicate rejection of the null hypothesis of non cointegration at the 10, 5,
and 1 percent significance levels, based respectively on critical values of 1.281, 1.644 and 2.326.

4.4. Panel cointegration estimations

We estimate the cointegrating vector using three methods: FMOLS, DOLS and PMG
estimators. The PMG computations were carried out using the Newton-Raphson algorithm in
a program written in Gauss12. Table 7 shows the saving retention coefficients of the three
models. There are interesting results. The saving retention coefficients estimated by FMOLS,
DOLS and PMG are respectively 0.38, 0.58 and 0.36 for the pool of all countries. In all cases,
the coefficients are statistically significant. This finding implies that capital is relatively
mobile in Africa. Our results are broadly consistent with earlier studies on the relationship
between saving and investment using panel data in Africa. Indeed, Adedeji and Thornton
(2006) report a savings-retention coefficient of 0.51 for DOLS estimate while that estimated
by the FMOLS is 0.73. Payne and Kumazawa (2005), using Pooled Ordinary Least Square,
Fixed effects and Random effect for a sample of 29 Sub-Saharan African countries, report that
during the period 1980-2001, the saving retention coefficients range between 0.20 and 0.24.
Using a panel of 36 African countries for the period 1980-2000, De Wet and Van Eyden
(2005) report a savings retention coefficient of 0.31, 0.34 and 0.28 respectively for Pooled,
fixed effects and random effects models.

There are some striking results when considering panels of country groups. There are marked
differences in retentions ratios between country groups, with the lowest being for non-oil
producing countries around 0.30 (FMOLS, DOLS and PMG) while the oil producing
countries exhibit a saving retention coefficient of 0.52 (FMOLS), 0.82 (DOLS) and 0.53
(PMG). The moderate degree of capital mobility in non-oil producing countries seems likely
to reflect the impact of non-market flows (foreign aid). The estimates for civil law countries
indicate a saving retention coefficient relatively high, 0.42 (FMOLS), 0,82 (DOLS), 0,44
(PMG) compared to common law countries with 0.30 (FMOLS), 0.32 (DOLS) and 0.29
(PMG). These results imply that in the countries with strong legal protections of investors
(common law countries), capital tends to be mobile internationally than in the countries with
worse protection (civil law countries).

The saving retention coefficients estimated by FMOLS, DOLS and PMG are respectively
0.46, 0.78 and 0.46 for the CFA Franc zone relatively higher compared to non CFA Franc
12
The program is available on http://www.econ.cam.ac.uk/faculty/pesaran.

12
zone 0.33 (FMOLS), 0.48 (DOLS), and 0.31 (PMG). On the one hand, this result shows that
CFA Franc zone cannot be viewed as an optimum currency area because capital mobility is
relatively low. On the other hand, the moderate degree of capital mobility in CFA franc zone
is due probably to the regional surveillance over macroeconomic policies. Since the 1994
devaluation, the CFA Franc countries have been guided by the example of the European
Union, which has used criteria for the public deficit and debt ratios to GDP as conditions for
countries to join EMU and then as ongoing conditions to be satisfied within EMU, according
to the previsions of the Stability and Growth Pact. Indeed, some authors as Coakley (1996)
argue that the high correlation between savings and investment rates simply reflects a binding
intertemporal budget constraint and current account solvency.

Table 7 : Panel cointegration estimation

Countries FMOLS DOLS PMG

All 0.38 (11.66) *** 0.58 (12.29) *** 0.36 (8.34) ***
FCFA 0.46 (8.78) *** 0.78 (27.34) *** 0.46 (6.60) ***
NCFA 0.33 (8.01) *** 0.48 (7.32) *** 0.31 (5.54) ***
OIL 0.52 (8.04) *** 0.82 (39.95) *** 0.53 (4.77) ***
NON OIL 0.30 (8.55) *** 0.29 (3.70) *** 0.33 (7.00) ***
FRENCH 0.42 (9.56) *** 0.85 (35.07) *** 0.44 (7.06) ***
ENGLISH 0.30 (6.36) *** 0.32 (4.33) *** 0.29 (4.74) ***
Notes: the value in parenthesis denotes the t-value for zero coefficients. *** significant at 1%.

5. Concluding remarks and policy implications

This paper has studied the international capital mobility of 37 African countries in terms of
the FH coefficient, applying recently developed panel cointegration methods. We apply three
classes of panel cointegration test and used PMG, FMOLS DOLS methods in order to deal
with heterogeneity problems and to conduct plausible tests.

The empirical findings reported in the paper reveal that savings and investment are
nonstationary and cointegrated series. Capital was relatively mobile in the African countries
in the sample during 1970-2006, with estimated saving retention coefficients of 0.38
(FMOLS), 0.58 (DOLS) and 0.36 (PMG) for the pool of all countries. There are some striking
results when considering panels of country groups. There are marked differences in retentions
ratios between country groups, with the lowest being for non-oil producing countries around
compared to that of oil producing countries. The same interpretation is true for common law
countries with savings retention coefficient relatively lower compared to civil law countries.
These results imply that in the countries with strong legal protections of investors, capital
tends to be mobile internationally than in the countries with worse protection. The more
striking result concern the CFA countries versus non CFA countries. The saving-retention
coefficient is relatively low in non CFA than in CFA countries implying that a monetary
union does not necessary conduct to better capital mobility.

The prevalence of moderate capital mobility has several implications for economic policies.
The observed moderate degree of capital mobility could be due to economic reforms and

13
structural adjustments, which are aimed at liberalization of markets, taking place in many of
these countries during the last two decades. Such reforms must be pursued.

The prevalence of moderate capital mobility of capital also implies that in these countries, the
prospects for economic growth need not to be severely constrained by the prevailing low level
of domestic savings.

References

Adedeji, O., & Thornton, J (2006). Saving, Investment and capital mobility in African
countries. Journal of African Economies, 16 (3), 393-405.
Banerjee, A., 1999. Panel data unit roots and cointegration: an overview. Oxford Bulletin of
Economics and Statistics, 61, pp. 607–629.
Bayoumi, T., (1990). Saving-Investment correlations: immobile capital government policy or
endogenous behaviour? International Economic Journal, 37, 360-387
Breitung, J., 2000. The Local Power of Some Unit Root Tests for Panel Data. In B. Baltagi
(ed.), Nonstationary Panels, Panel Cointegration, and Dynamic Panels, Advances in
Econometrics, 15, JAI, Amsterdam, 161-178.
Chen, B., McCoskey, S. & Kao, C. (1999). Estimation and Inference of a Cointegrated
Regression in Panel Data: A Monte Carlo Study, American Journal of Mathematical
and Management Sciences, 19, 75-114.
Coakley, J., J., Fuertes, A.M., & Spagnolo, F (2001). The Feldstein- Horioka puzzle is not as
bad as you think. Available from: http://papers.ssrn.com/sol3/papers.cfm.
Coiteux, M., & Olivier, S (2000). The saving retention coefficient in the long run and in the
short: evidence from panel data. Journal of International Money and Finance, 16, 535-
548.
Collier, P., & Gunning, J.W (2006). Why ha Africa grown so slowly. Journal of Economic
Perspectives, 13, 3-22.
De Wet, A.H & Van Eyden R.(2005). Capital mobility in Sub-Saharan Africa: A panel data
approach. South African Journal of Economics, 73, 1-22
Dooley, M., J. Frankel & Mathieson, D.(1987). International Capital mobility in developing
countries versus industrialized countries: What do savings-investment correlations tell
us? IMF Staff Papers, 34,503-529.
Feldstein, M., & Horioka, C (1980). Domestic saving and international capital flow. The
Economic Journal, 90, 314-329.
Haque, N., & Montiel, P. (1990). How mobile is capital in developing countries? Economics
Letters, 33, 359-362.
Ho, T (2002). The Feldstein-Horioka puzzle revisited. Journal of International Money and
Finance, 21, 555- 564.
Hurlin, C. & Mignon, V., (2005). Une synthèse des tests de racine unitaire sur données de
panel. Economie et Prévision, 3-4, (169), 253-294.
Hussein, K.A, Mello L.R (1999). International capital Mobility in developing countries:
theory and evidence. Journal of International Money and Finance, 18, 367- 381.
Im, K.S, M.H Pesaran & Shin, Y. C (2003). Testing for units roots in heterogeneous panels.
Journal of Econometrics, 115, 53-74.

14
Isaksson, A (2001). Financial liberalization, foreign aid, and capital mobility: evidence from
90 developing countries. Journal of International Financial Markets, Institutions &
Money, 11 (3), 309-338
Jansen, W.J (2000). International Capital mobility: evidence from panel data. Journal of
International Money and Finance, 19, 507- 511.
Journal of Econometrics, 115, 53-74.
Kao, C., & Chiang, M.H (2000). On the estimation and inference of a cointegrated regression
in panel data. Advances in Econometrics, 15, 179-222.
Kim, H., K-Y Oh & Jeong, C.W (2005). Panel cointegration results on international capital
mobility in Asian Economies. Journal of International Money and Finance, 24, 71-82.
Krol, T (1996). International Capital mobility: evidence from panel data. Journal of
International Money and Finance, 15, 467- 474.
La Porta R., F. Lopez-de-Silanes., A., Shleifer & Vishny, R.W (1998). Law and Finance.
Journal of Political Economy, 106, 1-53.
Levin, A., & Lin, C. F. (1993). Unit Root Tests in Panel Data: New Results. Discussion paper,
Department of Economics, UC-San Diego.
Levin, A., Lin, C., & Chu, C. (2002). Unit root test in panel data: asymptotic and finite
sample properties, Journal of Econometrics, 108 (1), 1-24.
Maddala, G.S., & Wu, S. (1999). A comparative study of unit root tests with panel data and
new simple test. Oxford Bulletin of Economics and Statistics, 61, 631-652.
Mamingi, N, (1997). Saving-Investment correlations and capital mobility: the experience of
developing countries. Journal of Policy Modeling, 19(6), 605-626.
Montiel J.P., (1994). Capital mobility in developing countries: some measurement issues and
empirical estimates. The World Bank Economic Review, 3, 327-342
Payne, J.E., & Kumazawa, R (2005). Capital mobility, foreign aid, and Openness: further
panel data evidence from Sub-Saharan Africa. Journal of Economic and Finance, 29,
122-126.
Pedroni, P., (1999). Critical values for cointegration tests in heterogeneous panels with
multiple regressors. Oxford Bulletin of Economics and Statistics, 61, 653-670.
Pesaran, M.H., (2005). A simple panel unit root test in presence of cross section dependence.
Cambridge University Working Paper.
Rocha, F (2000). Capital mobility in developing countries: evidence from panel data.
Department of Economics Working Paper. Universidad Sao Paulo, 1-15.
Sihna, T., & Sinha, D. (2004). The Mother of all puzzles would not go away. Economics
Letters, 82, 259-267
Stock, J., & Watson, M. (1993). A simple estimator of cointegrating vectors in higher order
integrated systems. Econometrica, 61, (4), 783-820.

15
Appendix

Table 1: Property of the data: descriptive statistics by country, 1970-2006

Country Investment rate Saving rate
Mean Std. Dev. Mean Std. Dev.
All countries 19.212 9.128 11.160 17.213
Algeria 30.439 6.527 35.008 8.490
Benin 16.053 3.374 1.631 4.206
Botswana 28.227 6.778 35.846 12.467
Burkina 18.901 3.324 2.765 5.509
Burundi 10.964 4.679 -2.019 7.563
Cameroon 19.767 7.322 19.919 4.281
Centrafrique 11.346 3.707 2.259 4.631
Chad 15.637 11.926 2.941 14.052
Congo. Dem. Rep. 10.770 4.985 9.675 5.308
Congo. Republic of 27.720 8.546 30.232 17.376
Cote d’Ivoire 15.501 6.206 21.362 5.651
Egypt 21.432 5.848 13.925 3.076
Gabon 31.905 10.131 48.911 11.673
Gambia 17.939 6.036 6.186 4.397
Ghana 14.945 7.996 7.295 3.429
Guinea-Bissau 23.953 9.399 -2.492 6.810
Kenya 18.640 2.234 16.448 4.847
Lesotho 38.663 15.527 -45.914 21.343
Madagascar 12.368 4.578 4.975 3.721
Malawi 17.966 5.281 9.037 6.711
Mali 19.049 4.679 4.838 6.632
Mauritania 21.932 13.424 2.713 14.470
Morocco 22.813 4.335 17.591 4.065
Niger 13.344 5.163 5.679 4.711
Nigeria 20.538 4.773 24.307 8.808
Rwanda 14.528 3.333 1.746 9.474
Senegal 18.200 4.715 7.181 3.755
Sierra Leone 10.100 3.502 6.749 10.386
South Africa 20.833 4.979 24.290 5.642
Sudan 13.545 3.920 8.485 5.087
Swaziland 22.527 7.106 11.902 14.325
Tanzania 16.832 5.439 4.432 4.193
Togo 19.261 4.433 13.136 11.725
Tunisia 25.569 3.854 22.731 2.154
Uganda 13.014 5.220 5.839 4.461
Zambia 18.386 8.109 17.776 12.486
Zimbabwe 17.246 3.545 15.559 5.721

16